We correlate the vertical component of ambient seismic noise data recorded on 56 broadband stations with dense coverage in the greater Los Angeles area to determine station-to-station Green’s functions. These Green’s functions provide an important test of community velocity models (Southern California Earthquake Center [SCEC] CVM 4.0 and CVM-H 5.2) used for strong ground-motion prediction for future scenario earthquakes in southern California. Comparisons of the ambient-noise Green’s functions for nearly 300 paths, with those calculated by the finite-element method in the community velocity models, reveal a strong waveform similarity for the dominant surface waves between 0.1 and 0.2 Hz. We find a mean correlation coefficient between the ambient-noise and finite-element Green’s functions of 0.62 for the CVM 4.0 and 0.49 for the CVM-H 5.2, indicating stronger waveform similarity for CVM 4.0. We also find that for 77% of the paths, the surface waves in the finite-element Green’s functions for CVM 4.0 arrive early, suggesting that the CVM 4.0 has velocities in the upper 10 km that are too fast along these paths. The same bias is evident for CVM-H 5.2, but is substantially smaller, with only 61% of the paths too fast. For 67% of the paths, CVM 4.0 has velocities faster than CVM-H 5.2. The time lags we obtain between the ambient-noise and finite-element Green’s functions provide key information for improving future community velocity models.